It’s Not You, it’s Me: Detecting Flirting and its Misperception in Speed-Dates

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 334–342, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP It’s Not You, it’s Me: Detecting Flirting and its Misperception in Speed-Dates Rajesh Ranganath Computer Science Department Stanford University rajeshr@cs.stanford.edu Dan Jurafsky Linguistics Department Stanford University jurafsky@stanford.edu Dan McFarland School of Education Stanford University dmcfarla@stanford.edu Abstract Automatically detecting human social in- tentions from spoken conversation is an important task for dialogue understand- ing. Since the social intentions of the speaker may differ from what is perceived by the hearer, systems that analyze human conversations need to be able to extract both the perceived and the intended social meaning. We investigate this difference between intention and perception by using a spoken corpus of speed-dates in which both the speaker and the listener rated the speaker on flirtatiousness. Our flirtation- detection system uses prosodic, dialogue, and lexical features to detect a speaker’s intent to flirt with up to 71.5% accuracy, significantly outperforming the baseline, but also outperforming the human inter- locuters. Our system addresses lexical fea- ture sparsity given the small amount of training data by using an autoencoder net- work to map sparse lexical feature vectors into 30 compressed features. Our analy- sis shows that humans are very poor per- ceivers of intended flirtatiousness, instead often projecting their own intended behav- ior onto their interlocutors. 1 Introduction Detecting human social meaning is a difficult task for automatic conversational understanding sys- tems. One cause of this difficulty is the pervasive difference between intended social signals and the uptake by the perceiver. The cues that a speaker may use to attempt to signal a particular social meaning may not be the cues that the hearer fo- cuses on, leading to misperception. In order to understand the impact of this dif- ference between perception and intention, in this paper we describe machine learning models that can detect both the social meaning intended by the speaker and the social meaning perceived by the hearer. Automated systems that detect and model these differences can lead both to richer socially aware systems for conversational understanding and more sophisticated analyses of conversational interactions like meetings and interviews. This task thus extends the wide literature on social meaning and its detection, including the detection of emotions such as annoyance, anger, sadness, or boredom (Ang et al., 2002; Lee and Narayanan, 2002; Liscombe et al., 2003), speaker characteristics such as charisma (Rosenberg and Hirschberg, 2005), personality features like ex- troversion or agreeability (Mairesse et al., 2007; Mairesse and Walker, 2008), speaker depression or stress (Rude et al., 2004; Pennebaker and Lay, 2002; Cohn et al., 2004), and dating willingness or liking (Madan et al., 2005; Pentland, 2005). We chose to work on the domain of flirtation in speed-dating. Our earlier work on this cor- pus showed that it is possible to detect whether speakers are perceived as flirtatious, awkward, or friendly with reasonable accuracy (Jurafsky et al., 2009). In this paper we extend that work to de- tect …

Content truncated to 500 words